The Flatiron Institute’s new supercomputer in New York City leads the newest Green500 List of the world’s most power-efficient supercomputers.
The latest list, which ranks supercomputers based on the number of floating-point operations per second (FLOPS) per watt of power, places the new supercomputer at 65.091 billion FLOPS per watt. This beats the previous record of 62.684 billion FLOPS per watt. The new Flatiron Institute supercomputer is the world’s most energy-efficient supercomputer, according to Green500.org, with 2.038 million billion FLOPS.
“This supercomputer opens up opportunities for doing new kinds of science. This is a workhorse machine, and we’re going to let our researchers try new things and drive discoveries,” Flatiron Institute’s Scientific Computing Core co-director Ian Fisk said.
The Flatiron Institute’s researchers will use the new supercomputer’s capabilities to solve complex computational astrophysics, biology, mathematics, neuroscience, and quantum physics challenges. The system, which uses NVIDIA‘s accelerated processing architecture, is well suited for machine learning applications such as multi-body simulations of the universe’s evolution, predicting how proteins fold and function, and discovering connections in genomic studies. Accelerated computing is also very good at linear algebra calculations, replicating how electrons behave at the quantum scale. This is because GPUs can perform far more calculations in parallel than traditional CPUs.
The system itself was constructed by Lenovo and is housed in a data centre in New Jersey. It exploits the efficiency characteristics of the company’s ThinkSystem SR670 V2, a server meant to be instal simply into standard data centres and be accessible to more researchers. Its efficient architecture keeps everything cool even in a high-heat, multi-GPU situation. The new Lenovo supercomputer incorporates 80 new NVIDIA H100 Tensor Core GPUs, linked via NVIDIA’s Quantum 200Gb/s InfiniBand network and housed in ordinary data centre racks.
“As a world leader in supercomputing, Lenovo is committed to making the same Exascale-ready technologies available for users of all sizes and across all industries. Our collaboration with Flatiron and NVIDIA demonstrates a pivotal milestone in enabling supercomputing technology at nearly any scale and supporting smarter technology for all,” Kirk Skaugen, Executive Vice President, Lenovo Group and President of Infrastructure Solutions Group said.
Researchers can start with a single server, progress to a partial rack, and eventually reach exascale levels, all while helping to improve scientific outputs and achieve new discoveries in sustainable computing.
The transistors in NVIDIA chipsets are only 5 nanometers wide, allowing for more computational muscle to be crammed onto the same size device. This decreased transistor size and numerous other modifications contributed to the new supercomputer’s lower energy consumption and higher performance efficiency.
There are numerous advantages to reducing a supercomputer’s power consumption without compromising performance. Electricity contributes significantly to the operational expenses and environmental impact of a machine. Every watt consumed generates heat, which must be evacuated from the system, necessitating even more power. Furthermore, more energy-efficient equipment can be installed in existing data centres without requiring costly electrical improvements.
“This computer will allow us to do more science with smarter technology that uses less electricity and contributes to a more sustainable future,” he says. “That’s what’s important to us,” Fisk said.
The new Flatiron Institute supercomputer has surpassed the Frontier Test & Development System (TDS) at Oak Ridge National Laboratory in Tennessee regarding power efficiency. Frontier TDS established a very high threshold on the June 2022 list, with 62.684 billion FLOPS per watt, while the main Frontier system — which broke the exaFLOP (1 billion FLOPS) barrier and remains the world’s fastest supercomputer — has 52.227 billion FLOPS per watt. As a result, unseating Frontier would be a difficult task, according to Fisk.
While both are record-breaking machines, the new Flatiron Institute supercomputer is a far different beast than Frontier, which was painstakingly conceived and built to achieve 1 exaFLOP. On the other hand, the new Flatiron Institute system uses widely available platforms, NVIDIA accelerated processing, and InfiniBand networking, making it “extremely high performance and very efficient without being particularly unusual,” according to Fisk. “It simply took a few of employees to get the system up and running.” This level of efficiency is now available to a far broader range of organisations, rather than simply the largest supercomputing centres.”
